棉纺织技术2025，Vol.53Issue(6)：58-65,8.
基于RAD-YOLO的织物疵点检测算法
Fabric defect detection algorithm based on RAD-YOLO
摘要
Abstract
To address the issues of diverse textile defect types,difficulty in recognizing small target,low algorithm precision and large number of parameters,making it difficult to deploy on edge devices,an improved fabric defect detection algorithm named RAD-YOLO based on YOLOv8n was proposed.Firstly,RFCADown module was designed,which reduced information loss during down-sampling through the integration of the receptive field coordinate attention within the down-sampling operation,efficiently enhancing the model's ability to learn global features.Secondly,the small target detection layer was added to the backbone network to optimize the algorithm and meet the model's demand for small defects recognition.Meanwhile,a C2f_DS module based on expanded residuals and spatial group attention was designed to optimize the multi-scale information extraction efficiency and strengthen the anti-interference ability of the model.Finally,a lightweight dynamic up-sampling module Dysample was introduced to preserve detailed information and further improve the accuracy of small target detection.The experimental results showed that the mAP@0.5 and mAP@0.5∶0.95 of the RAD-YOLO were 90.2%and 60.6%,respectively,which were 8.1 percentage points and 7.5 percentage points higher than those of YOLOv8n,and the model parameter amount was decreased by about 6.7%,and the model size was only 6.2 MB.The improved RAD-YOLO had better performance and could meet the needs of fabric defect detection.关键词
织物疵点检测/YOLOv8/感受野注意力/DWR/SGE注意力机制/DySampleKey words
fabric defect detection/YOLOv8/receptive-field attention/DWR/SGE attention mechanism/DySample分类
轻工业引用本文复制引用
王俊霖,方睿,刘金智,王畅阳,程鑫..基于RAD-YOLO的织物疵点检测算法[J].棉纺织技术,2025,53(6):58-65,8.基金项目
国家自然科学基金面上项目(62272067) （62272067）
国家重点研发计划项目(2020YFA0608000) （2020YFA0608000）
